WLAN QoS. Mathilde Benveniste a

Transcription

1 3 Mathilde Benveniste a With both the enterprise and residential sectors embracing voice over IP (VoIP) at an accelerating pace, and the pervasive use of wireless local area networks (WLANs), the natural requirement emerged for a technology to support VoIP over WLANs without degradation of its quality of service (QoS). QoS requirements for WLANs are imposed also by video and multimedia applications tailored for use with WLANs. A QoS-focused MAC Layer standard, IEEE e, was developed to meet the QoS requirements of a range of applications. In addition to VoIP/ multimedia QoS, the new standard serves mission-critical functions by reducing latency across a WLAN. This chapter discusses the enhancements the new standard adds to WLAN technology with respect to QoS performance, channel use efficiency, and power management of battery-based wireless devices. 3.1 Introduction Since the initial emergence of the network interface card for laptop computers and access points, the appeal of technology has been so strong worldwide that it is now appearing in a wide range of devices, including consumer electronics devices and VoIP phones. Enterprises wish to extend VoIP over wireless LANs for the convenience wireless service brings to the mobile user throughout the building, campus, quad and warehouse, as well as anywhere a WLAN is accessible. Residential users purchasing VoIP service for cost savings, look to the WLAN to enable them to make their telephones cordless. The installation of WLANs in public spaces, backed up by a ubiquitous Internet, makes the case of VoIP over WLANs even more compelling. Users can have telephone service portability free of any wires anywhere a WLAN is present. The new trends in the expansion of WLAN use include consumer electronics appliances generating multi-media traffic streams from applications such as video streaming and interactive gaming. All this could happen if wireless LANs could support QoS adequately in a congested WLAN was typically the reaction to the above observations prior to the adoption of the new standard for IEEE WLANs, known as IEEE e. The new standard enables frames from QoS-sensitive applications to be transmitted sooner than other frames, thus minimizing latency. It also introduces new power management features that will prolong the life of mobile devices powered by battery. The channel-use efficiency gains a Avaya Labs.

2 40 introduced by the new standard make it worth pursuing even in situations where all traffic is of the same type, thus allowing privileged treatment to none. The lower latency achievable with e enables also the prioritization of time-critical data. Devices observing the new standard can co-exist with compliant devices. This chapter gives a high level overview of the major mechanisms that have been modified and the new mechanisms introduced in e. They cover specifically the areas of channel access, admission control, and power management. QoS challenges that remain specifically in mesh networks are also discussed Terminology and Abbreviations For the reader reviewing the IEEE e standard [1], we note in this section some relevant naming conventions used in the standard. The QoS-aware contention-based random access is referred to as Enhanced Distributed Channel Access (EDCA). In the early standard drafts, and in much of the published literature, the same access approach had been called enhanced distributed coordination function (EDCF), following the naming convention of the standard, where Distributed Coordination Function (DCF) referred to contention-based random access [2]. Polled access was called Point Coordination Function (PCF) in the standard. The enhanced polled access mechanism in e is called HCF Controlled Channel Access (HCCA). An access point (AP) supporting e features is called a QAP, and a station equipped to use e features is called a non-ap QSTA. A QoS-aware WLAN, i.e., the group of stations supported by a QoS-aware AP, is called QBSS in the e standard as compared to BSS (basic service set), the group of stations supported by a legacy AP. Prioritization for the various functions of the channel access protocol is achieved by imposing waiting requirements of variable durations after the channel becomes available. The different durations are known as interframe spaces, with the shortest, SIFS (short interframe space), used when a transmission is acknowledged, PIFS (priority inter-frame space) used for PCF, and DIFS (DCF inter-frame space) required of stations using DCF. Abbreviations or acronyms used in this chapter are defined below. AP AIFS APSD CSMA/CA CW CWMax CWMin DCF DIFS EDCA HCCA PIFS PCF S-APSD SIFS access point arbitration inter-frame space automatic power save delivery carrier sense multiple access/collision avoidance contention window contention window maximum contention window minimum distributed coordination function DCF inter-frame space enhanced distributed channel access HCF controlled channel access priority inter-frame space point coordination function scheduled APSD short inter-frame space

3 41 TCMA TSPEC TXOP U-APSD WLAN tiered-contention multiple access traffic specification transmit opportunity unscheduled APSD wireless local area network 3.2 Channel Access A Wireless LAN operates on either the 2.4 GHz ISM band or the 5 GHz UNII band, each containing multiple radio channels. The IEEE standard specifies procedures for WLAN stations by which they share a single radio channel for asynchronous data transfer. Two channel access mechanisms are specified, contention-based and polled access. With contention-based access, stations transmit to peers and to the AP by accessing the channel using the distributed random access method that employs the CSMA/CA (Carrier Sense Multiple Access/Collision Avoidance) MAC protocol [3]. If an AP is present in a WLAN, peer-to-peer communication is not allowed independently of the AP. With polled access, the AP transmits frames to a station and polls for its transmissions. The IEEE e standard amendment also provides contention-based and polled access mechanisms, both of which represent enhancements of the mechanisms. The latter are referred to in this chapter as the legacy mechanisms e aims at reducing access delay and jitter in delivering QoS-sensitive frames from the source to the destination through enhanced functionality at the MAC Layer. The access delay comprises over-the-air time and queuing delay plus time consumed in retransmissions, when they occur. To achieve this goal, channel access in an e-compliant WLAN distinguishes among priorities of individual frames as introduced by IEEE 802.1D [4]. QoS-sensitive traffic is typically assigned higher priority than best effort data. Stations may transmit/ receive traffic streams of different priorities concurrently. The channel access mechanisms are described in detail later in this section. The e amendment also introduces features to improve channel use efficiency. It allows stations in WLAN served by an AP to communicate directly with one another. Signaling must be exchanged between the stations, through the AP, according to the Direct Link Setup protocol. A block acknowledgement mechanism is introduced in e, which improves channel efficiency by aggregating several acknowledgements into a single frame. Special signaling is needed between stations to negotiate this type of acknowledgement. Another efficiency enhancing feature introduced in e is the transmit opportunity (TXOP). In a contention based TXOP, a station may transmit a sequence of frames without having to contend for the channel, following a successful channel access attempt. Because all but one frame in a TXOP is transmitted without contention, TXOPs help reduce the frequency of collisions and thus increase channel use efficiency. A station granted a polled TXOP when polled by the AP may transmit several frames to the AP, thus obviating the need for additional polls. The remainder of this section describes the e channel access mechanism after first presenting the channel access protocols employed in , namely contention-based and polled access.

4 Legacy Channel Access Methods Legacy Contention-Based Channel Access According to the legacy contention-based access mechanism, DCF, each station listens to the channel and, if busy, postpones transmission and enters into the backoff procedure [2]. This involves deferring transmission by a random time, which facilitates collision avoidance between multiple stations that would otherwise attempt to transmit immediately after completion of the current transmission. The length of time for which a station will postpone its transmission depends on the backoff value, a number chosen randomly from a range of integers known as the Contention Window (CW). The backoff value expresses, in time slots, the cumulative time the channel must be idle before access may be attempted, excluding an additional time interval DIFS that the channel must remain idle following each period the channel is busy. In other words, transmission occurs after initially setting a backoff timer to the backoff value, and then counting down once for every slot of time that the channel remains idle following a busy period excluding an initial idle period of length DIFS. A transmission may not be attempted until the backoff timer expires. CW is set initially at the value CWmin, and is doubled after each collision involving its transmitted frame, until reaching the value CWmax; after which it remains constant for further retries. The frame is dropped if it cannot be transmitted successfully after a specified number of retries. CW is reset to CWmin following a successful transmission. When the backoff timer expires, it is reset to a new backoff value, drawn randomly from the contention window CW, regardless of whether there is a frame queued for transmission. Following anyone s transmission on the channel, a station is allowed to transmit only after the channel remains idle for at least a DIFS time interval. The value of DIFS is selected in order to enable DCF to share the same channel with the centralized protocol PCF of the standard. In PCF, the channel is accessed at the end of a transmission after a PIFS idle interval, which is shorter than DIFS Legacy Polled Channel Access At pre-specified regular time intervals, an AP engaged in polled access starts a contention free period by transmitting a beacon frame. The AP can access and maintain control of the channel, once the channel becomes idle for the duration of the contention-free period by transmitting after a shorter time interval, PIFS, than a station. In addition, a station hearing the beacon will refrain from transmitting if not polled until it receives notice from the AP that the contention-free period is over. The AP first transmits all broadcast and multicast frames and frames addressed to power-saving stations that subscribe to polled access. The transfer of frames to and from non-power-saving stations follows. The AP maintains a polling list of stations to be polled, and polls each of them at least once during a contention-free period. At the same time the AP transfers frames to the stations on the polling list. Every poll elicits a single frame from the polled station. The AP stops polling a station in a contention-free period if it receives a Null frame (frame containing a header but no body) in response to a poll. An

5 43 acknowledgement, data and/or poll can be combined into a single frame in order to save overhead e Contention-Based Channel Access The IEEE e contention-based access mechanism, EDCA, extends the contentionbased access mechanism of the standard to provide frame prioritization [1]. That is, given a collection of contending entities, prioritized access enables frames of higher priority to access the channel sooner. IEEE e uses the TCMA MAC protocol, a variant of CSMA/CA designed for priority differentiation [5, 6]. A key consideration in formulating EDCA was fairness. Because certain stations will transmit frames of different priorities, while others will transmit frames of a single priority, it was important for the channel access mechanism to provide the same performance to all frames of a given priority regardless of their source. Thus, instead of buffering all frames in a single queue (as with stations), an e station employs four queues, one for each access category based on the frame s priority [7]. The mapping of user priorities to access categories specified for a WLAN must be observed by all the stations. Traffic in an access category mapped to higher priorities will access the channel more readily than lower-priority access categories. The different queues of a station contend for channel access independently of one another, almost as if they resided in different stations. The only difference is that any internal collisions between two queues of the same station are resolved by allowing the higher priority frame to be transmitted, while the lower priority frame is treated as if it had experienced a collision. A contending queue in a station with multiple access categories behaves just like a station with traffic of a single access category with respect to accessing the channel. For simplicity of presentation, therefore, both are referred to as a station in the description of prioritized access that follows. The underlying MAC protocol in e contention-based access is CSMA/CA, which was described above. In e, the protocol s parameters CWmin and CWmax values are allowed to vary with the access category [8]. Assigning lower CWmin or CWmax values to an access category causes the contention window to be shorter when transmitting, or retransmitting following a collision. This indeed could offer priority differentiation, as shorter backoff values would lead to shorter average access delays, provided that the number of contending stations in the access category in question is small. The user should be cautioned, however, that when the number of contending stations in the access category is large, a short contention window would cause multiple stations to draw the same backoff, leading to collisions and consequently longer rather than shorter delays [9]. When differentiation with respect to CWmin or CWmax is pursued, the AP must be equipped to adjust these parameters dynamically in response to traffic conditions. The e standard permits such adjustments. Another priority differentiation mechanism for contention-based access is through differentiation of the arbitration time, referred to as AIFS in the e standard. This concept was introduced as part of the TCMA MAC protocol, which is described in the next section [5, 6].

6 44 Like the contending stations, the AP also differentiates between traffic of different priorities by using different access parameters. The access parameters used by the AP for a given access category may be different from those used by the stations. The AP is thus allowed to use higher-priority access parameters than the stations, a prudent measure since the AP typically transmits more traffic than a station. An important-efficiency enhancing feature introduced in e is the transmit opportunity [10]. Following a successful channel access attempt, a station may transmit a sequence of frames without having to contend for the channel. That is, a station is allowed consecutive transmissions of frames from the same access category without the need for backoff by using spacing between consecutive frames of SIFS, which is the shortest interframe spacing. The station thus maintains control of the channel for the entire TXOP by waiting a shorter time between transmissions than any other station. A good portion of a TXOP is also protected from collisions with hidden terminals. The acknowledgement to a frame indicates in the Duration field the length of the following frame in the TXOP or the remaining TXOP duration. This length is derived from the Duration field of the transmitted frame that is acknowledged. Because all but one frame in a TXOP is transmitted without contention, TXOPs help reduce the frequency of collisions. This increases channel use efficiency. Another useful efficiency measure introduced in e is the expiration of frames based on the time queued, waiting for transmission. The time limit for the expiration of a frame, known as MSDULifetime, varies with access category, as overly delayed frames may not be useful in some applications but useful in others. For instance, applications with low latency tolerance, like voice, drop excessively delayed frames on the receiving end. Excessively delayed frames are dropped in this case without further transmission attempts, thus making room for other transmissions. Naturally, one must be careful when setting MSDULifetime limits for different access categories to separate traffic types with different tolerance for packet loss. For instance, VoIP signaling and VoIP media should be in separate access categories in such a case. The impact of dropping excessively delayed frames has been studied in [5]. Figure 3.1 shows the effect of dropping voice frames if the time spent in the MAC layer exceeds the MSDULifetime limit. The average over-the-air delay experienced by a single station engaged in a voice call without dropping frames appears in Figure 3.1(a). Figure 3.1(b) shows the same resulting delay as a consequence of dropping any frames delayed by 20 milliseconds. Simulation Time (sec) Simulation Time (sec) Figure 3.1: Effect of dropping frames delayed in excess of MSDULifetime.

7 45 It is important to note that, because the contention-based access mechanism of e is based on access prioritization, part of the advantage it brings to a WLAN over the legacy CSMA/CA protocol is lost when the traffic load consists primarily of one type of traffic as for instance in the case of call centers, where much of the traffic comprises voice calls. The channel overhead penalty introduced by the longer MAC headers of e frames may counteract efficiency gains introduced in EDCA. The benefit of choosing EDCA over legacy DCF in such a case relates to the greater flexibility found in the former, as illustrated in Section TCMA MAC Protocol According to the CSMA/CA protocol, as implemented by , a station engaged in backoff countdown must wait while the channel is idle for time interval equal to DIFS before decrementing its backoff immediately following a busy period, or before attempting transmission. According to the TCMA (Tiered Contention Multiple Access) protocol, variable lengths of this time interval, which is called Arbitration-Time Inter-Frame Space (AIFS), lead to varying degree of accessibility to the channel [5,6]. A shorter AIFS will give a station an advantage in contending for channel access. Differentiation between different access categories is achieved by assigning a shorter AIFS to a higher priority access category. An example is shown in Figure 3.2. The effectiveness of priority differentiation of access categories is only partly due to allowing the station with the shortest waiting requirement to access the channel first, given two or more stations with expired backoff. This mechanism was used in to give priority to stations engaged in PCF to access the channel before any other station. For instance, an AP would wait an idle time period of length PIFS, which is shorter than the length of DIFS required of a station. When a legacy AP has to engage in backoff, however, it uses the same backoff countdown rules as a station. It must wait for an idle interval of DIFS duration before decrementing its backoff timer. Figure 3.2: AIFS differentiated contention-based access.

8 46 The benefit from shortening the waiting time for transmission is small relative to the effect of the different AIFS values when decrementing the backoff timer. Since countdown of the backoff timer following a busy period may not occur unless the channel has been idle for a time period equal to AIFS, backoff countdown of lower priority frames slows down, and even freezes, in the presence of higher-priority frames with expired backoff. This is because a transmission will occur and the channel will be busy again before the lower-priority station, with the longer AIFS, has a chance to decrement its backoff timer. This would occur commonly in congestion. Hence, in congestion conditions, the priority mix of stations with expired backoff timers favors higher priorities. In general, high priority stations will have lower backoff values than lower-priority stations when one looks at the residual backoff values of a mix of stations at any point in time. This desirable result is achieved without shortening the contention window from which the backoff value is drawn, which if pursued would increase the likelihood of collisions among the high-priority stations. Given any mix of initial random backoff values, the tendency of high-priority frames to reduce their backoff faster than lower-priority frames under TCMA leads to lower delay and jitter than without AIFS differentiation. Finally, the same tendency also reduces the likelihood of collisions between frames of different priorities, thus leading to a lower collision rate and higher throughput. These observations, which lead to the adoption of AIFS differentiation into the e standard, have been confirmed by subsequent performance evaluation studies [11 12] e Polled Channel Access The IEEE e standard improved the PCF polled channel access mechanism of the earlier standard to achieve better delay and jitter performance and greater channel use efficiency. The enhanced mechanism, called HCF controlled channel access (HCCA) in the e standard, resembles PCF, but with the following modifications. Polling is not limited to the contention-free period, but instead it can occur any time. The polling schedule is tailored to the time profile of the individual traffic streams, thus reducing both overhead, delay, and jitter. Overhead, delay, and jitter are also reduced through uplink TXOPs, which cause frames to be transmitted sooner than would have been possible otherwise. In general, a service period is a time interval of continuous communication between the AP and a station, comprised of downlink transmissions and/or a poll and the station s response to the poll. Polled-access service periods occur periodically at a negotiated service interval subject to limited time slippage. The AP transmits downlink frames to stations as single frames or as TXOPs. A downlink frame may be combined (or piggybacked) with a poll. With the poll, the AP grants a polled TXOP to the station. That means a response to a poll may consist of multiple uplink frames. An uplink frame can be combined with the acknowledgement to a downlink frame. The station can request extension of the TXOP by indicating the desired duration in a special QoS control subfield: TXOP Duration Requested. Uplink transmissions are protected from contention from other stations in the WLAN for the value of the Duration subfield in the downlink frame(s) sent to the station during the station s service period.

9 47 By allowing multiple frames to be transmitted uplink without contention, in response to a single poll, a lot of the signaling frames that would otherwise be required are eliminated. TXOPs reduce contention when employed by either access method. TXOPs that are secured by the AP and granted to a station employing polled access give the station priority over any station using contention-based access, regardless of their respective priorities. To match polling frequency to the traffic, a station that starts a new traffic stream exchanges signaling with the AP to establish the schedule by which the station will be polled. A station may have several traffic streams going on at once. An ADDTS frame is submitted for each traffic stream associated with the station, describing various aspects of transmission/delivery in the TSPEC element. These include the following: the nominal size of data frames (Nominal MSDU Size), the average bit rate at which data is generated (Mean Data Rate), the maximum delay allowed for queuing and transport of frames across the channel (Delay Bound), the maximum time allowed between consecutive service periods granted to the station (Maximum Service Interval) for the traffic stream, and the minimum physical bit rate to be assumed in establishing a schedule (Minimum PHY Rate). Each stream may have a different polling schedule. Alternatively, a station may request a single aggregate polling schedule for all admitted traffic streams. It does so by setting the Aggregation subfield in the TS Info Field of the TSPEC element equal to 1. If the AP can accommodate the stream specified in the ADDTS request, it will indicate so in an ADDTS response that includes the Schedule element, specifying the schedule of the delivery of data and polls. If an ADDTS request is declined, the station may employ contention-based access for the traffic stream. A traffic stream is deleted when a station sends a DELTS frame to the AP. The negotiation between the station and the AP in establishing a polling schedule for each traffic stream, through the submission of ADDTS frames, provides a stand-alone admission control mechanism. As explained above, polled access has priority over contention-based access. It is not necessary, therefore, to restrict access of coexisting contention-based stations through admission control in order to enable polled stations to enjoy guaranteed delay/jitter performance. The enhanced polled access mechanism of the e standard may operate during both the contention and the contention-free periods into which the channel time is typically partitioned. The AP can access the channel during the contention period by using PIFS, a shorter waiting requirement than that for stations, to initiate service periods for the stations with admitted traffic streams [13 15]. As a consequence, it is expected that, in practice, e APs will allocate most of the channel time to contention periods. Compared to the legacy PCF mechanism, the e polled access mechanism results in a polling schedule that better matches the generation of frames in a periodic traffic stream. This results in superior delay/jitter performance and better channel use efficiency. The transmission of multiple uplink frames per poll also increases channel use efficiency. Relative to contention-based access, scheduled polled access leads to better channel use efficiency because stations in the same WLAN (that is, stations served by the same AP) do not contend for the channel, thus eliminating the possibility of collision among them. The superior delay/jitter performance of polled access in e makes it the ideal choice for voice and streaming multimedia applications.

10 Illustrative Examples Time-sensitive traffic occurs in diverse environments, with a different mix of traffic priorities. The prioritization capability of EDCA has been demonstrated in several performance studies [5, 9, 12]. Figure 3.3 illustrates the impact of AIFS differentiation on the average over-the-air delay experienced by nine high-priority voice streams using an b channel in the presence of lower priority data traffic, considered in [5]. Figure 3.3(a) shows the average delay experienced by the voice streams if the legacy DCF access mechanism was used and Figure 3.3(b) shows the delay experienced with EDCA. Simulation Time (sec) Simulation Time (sec) Figure 3.3: Average delay for top priority traffic category. Prioritized access is useful if both low and high priority traffic are present in the same WLAN. The question thus arises whether EDCA would be of value in WLANs carrying mostly traffic of the same priority, such as call centers. The value of EDCA in such environments stems from its flexibility and the efficiency of channel utilization it introduces. For instance, EDCA can be of benefit because it allows the AP to use different access parameters than the stations. Identical EDCA access parameters across all entities contending in a given priority class lead to consistent performance for all the traffic in that priority class only if these entities have comparable traffic loads. There is a pronounced load-induced inequity in the case of the AP. The AP has more traffic to transmit than any individual station since the uplink traffic is distributed among multiple stations and, in general, the downlink traffic in a WLAN is heavier than the uplink traffic. In the case of voice calls, the AP must transmit multiple voice streams, one for each station engaged in a voice call, while the stations transmit one voice stream each. By allowing the AP to contend for the channel with higherpriority EDCA parameters, downlink delays are shortened and become comparable to those of uplink voice streams. Allowing the AP to access the channel with a shorter AIFS duration than the stations and no backoff requirement increases the voice capacity of a WLAN by as much as 38 per cent [16]. The voice capacity of a WLAN is the number of simultaneous voice calls that result in bounded over-the-air delays and no buffer overflow. Assuming an error-free channel, 46 voice calls with 20 milli-second frame interarrival time can be carried in an a WLAN when the AP uses the same access parameters as the stations. The WLAN voice capacity becomes 58 when the AP is allowed to transmit with AIFS equal to PIFS and a contention window of size zero. 1 1 Since the conference proceedings where the results in reference 16 of this chapter are not readily available, a synopsis is included as an endnote.

11 49 Using the HCCA polled access mechanism of e can increase the voice call capacity of a WLAN further, as it provides collision-free transmission. For comparable conditions, the voice capacity of an a WLAN is 65 voice calls [17]. The use of HCCA introduces a 12 per cent increase relative to the capacity achieved with optimized AP access parameters. Such a gain may seem insufficient to justify the complexity of implementing the scheduling algorithm required for HCCA. Considering EDCA as the alternative, some algorithmic complexity is also needed in order to achieve high capacity consistently. It relates to the choice of the access parameter values for different traffic conditions. The e standard does not specify how these parameters must be set; a task left to the user. The appropriate choice of a certain EDCA parameter value namely, the contention window size depends on the traffic conditions. The wrong choice could result in capacity loss, because of aggressive behavior and a high collision rate. This point is illustrated in [17], where choosing the standard default contention window value leads to a capacity of only 35 voice calls for an a WLAN employing EDCA. In the absence of special optimization algorithms for adaptation of the contention window size to traffic intensity, this EDCA parameter should be assigned a large fixed value. Naturally, assigning large fixed values to the EDCA contention window size removes its effect on prioritization, leaving the AIFS size as the main priority differentiator between traffic classes. While this works for exclusively e WLANs, mixed systems are problematic. The range of ten time slots provided in the e standard for the AIFS duration is sufficiently wide to enable adequate priority differentiation [11, 12]. The entire range is not available, however, when e-compliant stations must co-exist with legacy stations. Legacy stations, which employ an AIFS interval of length DIFS+1, must be treated as having low priority traffic. The effective AIFS range is thus reduced to a single time slot, which may not be sufficient for differentiation among multiple classes. Additional differentiation would thus be of value. Hence the need to differentiate based on contention window size in this special case, in spite of the caveats. 3.3 Admission Control Admission control provides bandwidth management to ensure that QoS-sensitive applications, such as voice and video, will be afforded a satisfactory quality of service. Overloading the WLAN with an excessive number of users entitled to high-priority access would make it hard to provide consistent QoS. Therefore, requests are submitted by stations for the admission of specific traffic streams to the AP, which keeps track of the traffic on the channel and accepts or declines the request. The information contained in this exchange will depend on the channel access method involved; it will be different for contention-based access and for polled access. Admission control is an intrinsic part of polled access, and thus comes automatically with the decision to use this access method. Admission control is an option that is available for stations using contention-based access. It is important to note that admission control becomes imperative for contention-based stations with QoS traffic in a WLAN that supports polled access, unless polled access is limited to just top priority traffic. Stations using contention-based access will access the channel with lower priority than any station

12 50 that uses polled access, regardless of their respective traffic priorities. Because the AP can transmit before any station, it can give a polled station an opportunity to transmit before any contention-based station Admission Control for Contention-Based Channel Access Admission control for contention-based access is an optional feature for a station and an AP. It involves the decision at the AP to allow stations that employ contention-based access in the WLAN to transmit traffic using the parameters of an access category. This enables the AP to track and manage bandwidth use. It is not necessary to impose admission control on all access categories. The e contention-based access mechanism shields the admitted traffic from contention by lower-priority transmissions. It is important, however, to require admission control in all access categories of higher priority than the access category of the traffic of interest. The contention to be experienced by traffic in a given access category cannot be bounded if traffic in access categories of higher priority is unrestricted. The basic procedure of admission control for distributed access is the following. In its beacons, the AP advertises to the WLAN the access categories that are protected by admission control. A station that has traffic to transmit or receive in a protected access category must request permission from the AP before it is allowed to do so. The signaling is similar to that used for the admission of a traffic stream for polled access [18]. A station s request, submitted in an ADDTS (add Traffic Specification) frame, describes the traffic stream to be admitted. The description includes the data frame size, the mean data rate, and the minimum physical transmission rate for each of the directions on which the channel would be accessed with the parameters of the access category in question. If an ADDTS frame indicates a bi-directional traffic stream, traffic is specified for one of the two directions; the other is assumed to be the same. The response to the ADDTS request, if affirmative, furnishes in the Medium Time field the channel time the station is allotted for uplink transmissions using the parameters of the access category specified in the request. The allotted channel time is expressed as the number of time units the channel may be used by the station for its transmissions over a fixed known time interval. If the AP declines an ADDTS request, the station may still transmit, but with parameters of a lower-priority access category that requires no admission control. There should be at least one access category without the admission control requirement. Stations that do not support admission control may transmit only with parameters of access categories of equal or lower priority, and for which admission control is not mandatory. Once a station receives its allotted channel time for a particular access category, it keeps track of the portion that has been used up for its transmissions, and for any retransmissions. The station may request additional channel time for an admitted traffic stream if its allocation is being used up too fast, or if a new data flow is added to the same traffic stream. A single admitted traffic stream could be specified per access category, which would be the aggregate of several data flows. The station updates the combined requirements of all data flows in the access category in question and sends a new ADDTS request for an updated allocation. To give up all of its allotted channel time for a particular access category, a station submits a DELTS (delete Traffic Specification) frame. The

13 51 channel time allotted to a station for an access category is released if no transmissions in that access category to/from the station have occurred for a specified time period, the Inactivity Interval, which is indicated on the ADDTS frame Admission Control for Polled Channel Access Admission control is exercised automatically when using polled channel access. The AP will reject an ADDTS request if it cannot meet the requirements for a service period schedule requested by a station for a traffic stream. If the requested requirements can be met, the AP responds with a service period schedule. Unlike in the case of contentionbased access, a station using polled access may have several admitted traffic streams of the same priority. During the negotiation, a minimum set of parameters must be specified in the ADDTS request so that the AP can schedule time on a service period for the traffic stream that is to be admitted. These parameters include mean data rate, frame size, minimum transmission rate, and either the maximum service interval or a delay bound. If a traffic stream is admitted, the ADDTS response will include non-zero values for mean data rate, frame size, minimum transmission rate, and the maximum service interval. The ADDTS response will include a Schedule element, which provides the schedule of the delivery of data and polls. The minimum transmission rate will be used in determining the length of TXOPs and service periods. The priority of a traffic stream may be considered in admission control. An admission control request from traffic stream with a higher priority may cause an admitted stream to be dropped. The AP sends a DELTS frame to notify a station that a traffic stream is dropped. Admission of a traffic stream may therefore not be guaranteed. 3.4 Power Management Several of the QoS-sensitive applications will involve multimedia traffic over batterypowered handheld devices, such as a PDA or a wireless VoIP phone. In crafting a standard of good QoS performance, it was thus considered important to prolong the battery life of such devices. The e standard amendment offers several new mechanisms to help battery-powered devices conserve power by enabling them to power down their receivers and transmitters intermittently without losing connectivity or data. The new power management mechanisms apply to WLANs served by an AP such WLANs are known as infrastructure WLANs, and for this reason, the discussion in this section will focus on power saving methods for infrastructure WLANs. A station informs the AP of its operating power-management mode, power saving versus active, when it associates with the WLAN. The mode can be changed during the association period by changing the Power Management bit, a bit in the frame control field of the frames transmitted by the station. The AP will not send transmissions to a station that has declared itself to be in power save mode, unless it knows that the station has its receiver fully powered, i.e., it is in the awake state, and ready to receive. Otherwise, the AP will assume that the station s receiver is powered down, i.e., it is in the doze state, and

14 52 for this reason, any incoming frames addressed to a power-saving station will be buffered for later transmission. A simple, but not efficient, way for a power-saving station to retrieve multiple buffered frames at once is to switch its power management mode to active. A data frame, or a Null frame sent by station to the AP with the Power Management bit set to 0 will enable the AP to transmit the buffered data. The station may subsequently return to powersave mode using another frame with the Power Management bit set to 1. The inefficiency in this approach stems from the fact that it requires extra frames to be transmitted for signaling purposes e introduces delivery methods with reduced signaling. The AP may deliver buffered frames to their destination power-saving stations either on a previously negotiated schedule or in response to transmissions from the respective stations that initiate such delivery. In order to initiate delivery in the latter case, a station should know that there are frames for it buffered at the AP. Notification of the presence of buffered frames at the AP typically comes through a special station-specific field contained in the beacon frames broadcast by the AP, or in reserved fields of downlink frames directed to the individual stations. In some situations, as we will see, notification is not provided by the power-saving mechanism, and thus it must be furnished either by the application running on the station, or by transitioning to a different power saving mechanism that provides such notification. The station chooses the delivery and notification mechanisms and communicates it to the AP either upon association or re-association of the station with the WLAN or through explicit signaling using an ADDTS frame. The various mechanisms available in an infrastructure WLAN will be described in the following section. They include (1) the legacy power save mechanism, which was available pre e. APSD (automatic power save delivery) was introduced by the e standard to reduce the signaling that would otherwise be needed for delivery of buffered frames to power-saving devices by an AP. APSD provides two ways to start delivery: (2) scheduled APSD (S-APSD) and unscheduled APSD (U-APSD). Unscheduled APSD can take (3) a full U-APSD or (4) hybrid U-APSD form. With full U-APSD, all types of frames use U-APSD independently of their priority. Hybrid U-APSD employs a combination of U-APSD and the legacy power save mechanism Legacy Power-Save Mechanism The legacy power-save mechanism applies to both infrastructure WLANs and WLANs without an AP. We describe here how it works with the former since the new power save mechanisms deal only in WLANS served by an AP. For information on the latter, the reader is referred to the standard. Frames buffered at the AP for a power-saving station employing contention-based access are delivered when the station sends a special control frame, the power save poll (PS-Poll). The AP sends a single buffered frame to a station after receiving a PS-Poll, either immediately or soon thereafter. More PS-Polls are required in order to retrieve additional buffered frames. The presence of further frames remaining at the AP is indicated by the More Data bit of the control frames of the transmitted frame, which is set to 1. A station using legacy power save can rely on the traffic indication map (TIM) to learn if the AP holds buffered data for it. The TIM is a bit map containing the buffer status

15 53 per destination station. It is sent regularly on beacon frames broadcast by the AP at known times. If the station is in the doze state, it will wake up at the beacon times to receive and interpret the TIM. Alternatively, a station can ascertain the presence of additional frames buffered for it at the AP while receiving a buffered frame. The More Data bit in the control field of that frame would have been set to 1 if additional frames remained buffered for the station. A power-saving station that supports legacy polled access need not send PS-polls in order to receive its buffered frames. The station receives its buffered frames at the start of the contention-free period, when it awakens to listen to the TIM and learn of its buffer status at the AP. Such a station would probably not request to be on the polling list because that would require staying awake for the entire contention-free period. Uplink frames are sent by contention in that case Automatic Power Save Delivery APSD is a mechanism for the delivery of unicast frames from the AP to a power-saving station. This mechanism was introduced by e in order to reduce the signaling traffic caused by PS-Polls and their acknowledgements. A station may use both APSD and legacy PS-Polls at the same time to retrieve buffered frames from the AP. Certain restrictions apply, however, which are discussed below. To use APSD, stations must have the Power Management subfield in the control field of all transmitted frames set to 1. The AP may deliver buffered frames to their destination power-saving stations either on a previously negotiated schedule or in response to receiving transmissions from the respective stations that trigger such delivery. The two APSD approaches are thus known as scheduled and unscheduled. A station may use both approaches at the same time, provided that only one is used for a given access category Scheduled APSD This mechanism is well suited for periodic traffic streams, such as voice and audio/video, and is especially good for unidirectional downlink periodic streams. With scheduled APSD, downlink transmissions to power-saving devices will occur at a schedule that is known in advance, obviating the need for special signaling between the station and the AP. The AP and the station negotiate in advance a time schedule by which the station will power its receiver fully to receive any frames that are buffered for it at the AP. A station that wishes to use S-APSD must send an ADDTS request with the APSD subfield in TS Info field of the TSPEC element set to 1. The TSPEC element contains the time of the first downlink transmission (Start Service Time) as well as the time interval at which downlink transmission will be repeated (Service Interval), as in the case of polled access. The Start Service Time is expressed in terms of the time shared in the WLAN, known as the TSF timer [19]. While the Start Service Time field is used optionally with polled access, this field must be specified when using Scheduled APSD, as knowledge of the time of downlink frame delivery affords a station the longest stay in the doze state. The AP is given the last say in setting the start time of the periodic transmissions to the station so that its transmissions to different power-saving stations are staggered in a way that minimizes the time the power-saving stations are awake. If the request is

16 54 accepted, the AP will return an ADDTS response containing a Schedule element, which, among other, includes the Start Service Time selected by the AP. The station will wake up to receive its buffered frames at the times indicated by the returned schedule. Either channel access method, polled or contention-based access can be used with Scheduled APSD. Scheduled APSD fits naturally with polled access. To indicate polled access, the Access Policy subfield of the ADDTS TS Info field would be set to (0, 1), and the Start Service Time field in the TSPEC element must have a nonzero value. When the station plans to use contention-based access with Scheduled APSD, the Schedule subfield of the ADDTS TS Info field must be set to 1, and the Access Policy subfield must be set to (1,0). For stations using Scheduled APSD in conjunction with contention-based access, the uplink transmissions do not require polling. A power-saving device that uses contentionbased access can transmit to the AP at any time Full Unscheduled APSD Unscheduled APSD was introduced for stations accessing the channel by contention, in order to enhance the efficiency of legacy power save. A power-saving station may use not just a PS-Poll, but also any data or Null frame referred to as a trigger frame in order to notify the AP that its receiver is fully powered and ready to receive transmissions [20, 21]. Using a data frame that is pending transmission at the station, instead of a PS-Poll, to initiate downlink transmission clearly reduces the traffic generated by the station and increases battery life and channel use efficiency. Additional gains are achieved from relaxing the number of frames the AP is allowed to transmit to a power saving station when it receives notice to do so. While receiving a PS- Poll from a station allows the AP to transmit a single downlink frame -- of the highest priority access category buffered -- receiving a trigger frame will start an APSD service period for that station. During a service period, the AP may send multiple frames, subject to a limit specified by the station. Eliminating the extra signaling that would otherwise be necessary under legacy power save also increases the efficiency of channel use and conserves battery life. Naturally, since the station does not know in advance the number of frames sent by the AP in a service period, it must be notified when the last frame has been transmitted for a given service period so it may transition to the doze state. The control subfield EOSP in the last delivered frame marks the end of a service period. The AP need not deliver all frames buffered for a station in a single service period. As in the case of legacy power save, the More Data control subfield in a last frame transmitted in a service period indicates whether there are frames remaining buffered at the AP. Knowing its AP buffer status enables the station to send another trigger frame or PS-Poll to retrieve more of its buffered frames. As with the legacy power-save mechanism, a station can learn about its buffer status by listening to the beacons for its TIM [21]. This is needed only while not receiving frames from the AP, as the More Data control subfield in downlink frames to the station conveys the same information. To use full U-APSD, a station sets the first four bits of the QoS Info subfield of the QoS Capability element in the (re-) association request all to 1. The Max SP Length